Automating Discovery of Software Tuning Parameters

摘要

Software Tuning Panels for Autonomic Control (STAC) is a project to assist in the integration of existing software into autonomic frameworks. It works by identifying tuning parameters and rearchitecting to expose them as a separate control panel module. The project poses three distinct research challenges: automating the identification of tuning parameters, rearchitecting to centralize and expose them, and combining these two capabilities to facilitate the integration of existing software into autonomic frameworks. Our previous work focused on the second problem, automating the rearchitecture to expose and isolate tuning parameters. In this paper we concentrate on the first problem, automating the identification of tuning parameters. We begin with an empirical study of documented tuning parameters in a number of open source applications. Prom our observations of these known tuning parameters, we create a catalogue of different kinds and organize them into a taxonomy. Finally, we characterize a member of the taxonomy as a source code pattern that is used to find similar tuning parameters. We report our experience in applying this methodology in the context of a large, open source Java?system.